A vital material property of metals is long life fatigue strength. It describes the maximum load that can be cyclically applied to a defined specimen for a number of cycles that is thought to represent an infinite lifetime. The experimental measurement of long life fatigue strength is costly, justifying the need to create a precise estimate with as few experiments as possible. We propose a new approach for estimating long life fatigue strength that defines a ready-to-use experimental and analysis procedure. It relies on probabilistic machine learning methods, efficiently connecting expert knowledge about the material behavior and the test setup with historical and newly generated data. A comparison to state-of-the-art standard experimental procedures shows that our approach requires fewer experiments to produce an estimate at the same precision—massively reducing experimental costs.